本文将介绍两台Mysql如何实现高可用架构。通常我们会配置主从同步，但这样若主的Mysql挂掉，还需要手动干预，例如把指向主库的IP地址修改为指向从库的IP，为了实现自动切换到从数据库，我们可以使用Keepalived配置一个浮动的VIP出来供前端访问，主的Mysql故障了VIP会立即自动切换到从的Mysql，省去了人工干预的时间，但要想故障的那台Mysql起来后也能作为从Mysql自动去同步数据，就需要配置成互为主从。拥有VIP的Mysql可认为是主库，才能进行数据的写入，由于VIP可以在两台Mysql之间浮动切换，因此这两台mysql是互为主从。

一、测试环境

操作系统版本：Red Hat Enterprise Linux Server release 6.5 (Santiago)

Mysql版本：MySQL-5.6.38-1.el6.x86_64.rpm-bundle.tar

keepalived版本：keepalived-1.2.7-3.el6.x86_64.rpm

node01：192.168.10.71

node02：192.168.10.72

VIP：192.168.10.70

二、配置node01为主、node02为从的主从同步

1、node01和node02分别安装好Mysql 5.6.38，安装方法请参考上一篇博文。

2、在node01和node02分别编辑/etc/my.cnf，配置如下：

[mysqld]

log-bin=mysql-bin

server-id = 1

[mysqld_safe]

log-error = /var/log/mysqld.log

pid-file = /var/run/mysqld/mysqld.pid

replicate-do-db = all

3、重启mysql服务

[root@node01 ~]# service mysql restart

[root@node02 ~]# service mysql restart

4、登录node01的mysql，创建用于同步的账户repl，密码为123456，并查询master状态，记下file名称和posttion数值

mysql> GRANT REPLICATION SLAVE ON *.* to 'repl'@'%' identified by '123456';

Query OK, 0 rows affected (0.00 sec)

mysql> show master status；

5、登录node02的mysql，执行以下语句开启从服务器，注意master_host要填写node01的IP

mysql> change master to master_host='192.168.10.71',master_user='repl',master_password='123456',master_log_file='mysql-bin.000001',master_log_pos=318;

Query OK, 0 rows affected, 2 warnings (0.36 sec)

命令参数解释：master_host='192.168.10.71' ## Master 的 IP 地址master_user='repl' ## 用于同步数据的用户（在 Master中授权的用户）master_password='123456' ## 同步数据用户的密码master_log_file='mysql-bin.000001' ##指定 Slave 从哪个日志文件开始读复制数据（可在 Master 上使用 show master status 查看到日志文件名）master_log_pos=429  ## 从哪个 POSITION 号开始读

mysql> start slave;

Query OK, 0 rows affected (0.03 sec)

6、查询从服务的状态，状态正常

7、在node01创建一个数据库、一个表并插入一行数据，用于测试node02是否能同步过去

mysql> create database mysql_long;

Query OK, 1 row affected (0.00 sec)

mysql> use mysql_long;

Database changed

mysql> create table test(id int(3),name char(5));

Query OK, 0 rows affected (0.13 sec)

mysql> insert into test values (001,'jlong');

Query OK, 1 row affected (0.01 sec)

8、登录到node02的Mysql，同步正常

三、配置node02为主、node01为从的主从同步

1、登录node02的mysql，创建用于同步的账户repl，密码为123456，并查询master状态，记下file名称和posttion数值，并查询master状态

mysql> GRANT REPLICATION SLAVE ON *.* to 'repl'@'%' identified by '123456';

Query OK, 0 rows affected (0.00 sec)

Query OK, 0 rows affected (0.10 sec)

2、登录node01的mysql，执行以下语句开启从服务器，注意这里master_host要填写node02的IP

mysql> change master to master_host='192.168.10.72',master_user='repl',master_password='123456',master_log_file='mysql-bin.000001',master_log_pos=318;

Query OK, 0 rows affected, 2 warnings (0.36 sec)

mysql> start slave;

Query OK, 0 rows affected (0.03 sec)

3、查询从服务的状态，状态正常

4、在node02创建一个数据库、一个表并插入一行数据，用于测试node01是否能同步过去

mysql> create database mysql_long2;

Query OK, 1 row affected (0.00 sec)

mysql> use mysql_long2;

Database changed

mysql> create table test2(id int(3),name char(10));

Query OK, 0 rows affected (0.71 sec)

mysql> insert into test2 values (001,'jianlong');

Query OK, 1 row affected (0.00 sec

5、node01同步正常。

这样互为主从就配置好了，两台机既是对方的Master，又是对方的Slave，无论在哪一台机上数据发生了变化，另一台都能及时进行同步数据，下面我们开始配置keepalived实现高可用。

四、keepalived配置

1、在node01上使用yum安装keepalived

[root@node01 ~]# yum install keepalived -y

2、在node02上也使用yum安装keepalived

[root@node02 ~]# yum install keepalived -y

3、编辑node01的keepalived的配置文件

[root@node01 ~]# cat /etc/keepalived/keepalived.conf  ! Configuration File for keepalived global_defs {   notification_email {     acassen@firewall.loc     failover@firewall.loc     sysadmin@firewall.loc   }   notification_email_from      Alexandre.Cassen@firewall.loc   smtp_server 192.168.200.1   smtp_connect_timeout 30   router_id LVS_DEVEL} vrrp_instance VI_1 {    state BACKUP   ##node01和node02都配置成BACKUP，角色由优先级确定    interface eth2    virtual_router_id 71    priority 100      ##node01的优先级设置比node02的高    advert_int 1    nopreempt        ##设置不抢占（需在BACKUP状态下设置才有效）    authentication {        auth_type PASS        auth_pass 1111    }    virtual_ipaddress {        192.168.10.70    }} virtual_server192.168.10.70 3306 {    delay_loop 6    lb_algo wrr    lb_kind DR    nat_mask 255.255.255.0    persistence_timeout 50    protocol TCP     real_server 192.168.10.71 3306 {        weight 100        notify_down /etc/keepalived/stopkeepalived.sh #3306端口不可用则执行脚本        TCP_CHECK {        connect_timeout 10        nb_get_retry 3        delay_before_retry 3        connect_port 3306                    }    }}

4、编辑node02的keepalived的配置文件

[root@node02 ~]# cat /etc/keepalived/keepalived.conf  ! Configuration File for keepalived global_defs {   notification_email {     acassen@firewall.loc     failover@firewall.loc     sysadmin@firewall.loc   }   notification_email_from      Alexandre.Cassen@firewall.loc   smtp_server 192.168.200.1   smtp_connect_timeout 30   router_id LVS_DEVEL} vrrp_instance VI_1 {    state BACKUP       ##node01和node02都配置成BACKUP，角色由优先级确定    interface eth3    virtual_router_id 71    priority 90        ##node02的优先级设置比node01的低    advert_int 1    nopreempt          ##设置不抢占（需在BACKUP状态下设置才有效）    authentication {        auth_type PASS        auth_pass 1111    }    virtual_ipaddress {        192.168.10.70    }} virtual_server192.168.10.70 3306 {    delay_loop 6    lb_algo wrr    lb_kind DR    nat_mask 255.255.255.0    persistence_timeout 50    protocol TCP     real_server 192.168.10.72 3306 {        weight 100        notify_down /etc/keepalived/stopkeepalived.sh #3306端口不可用则执行脚本        TCP_CHECK {        connect_timeout 10        nb_get_retry 3        delay_before_retry 3        connect_port 3306                    }    }} [root@node02 ~]#

5、node01和node02都编辑一个keepalived的自杀脚本/etc/keepalived/stopalived.sh，一旦检测到Mysql的3306端口不通，便执行此脚本触发vip的切换，脚本内容很简单，就是service keepalived stop就行，因为keepalived服务停止便会触发高可用的切换动作。

6、启动keepalived服务

[root@node01 ~]# service keepalived start

Starting keepalived:                                       [  OK  ]

[root@node02 ~]# service keepalived start

Starting keepalived:                                       [  OK  ]

7、观察node01的message日志，由于node01的优先级高，因此进入了master角色，vip也已经加上了

[root@node01~]# tail -f /var/log/messagesOct 30 22:47:46node01 Keepalived[3170]: Starting Keepalived v1.2.7 (09/26,2012)Oct 30 22:47:46node01 Keepalived[3171]: Starting Healthcheck child process, pid=3173Oct 30 22:47:46node01 Keepalived[3171]: Starting VRRP child process, pid=3174Oct 30 22:47:46node01 Keepalived_vrrp[3174]: Interface queue is emptyOct 30 22:47:46node01 Keepalived_vrrp[3174]: Netlink reflector reports IP 192.168.10.71 addedOct 30 22:47:46 node01 Keepalived_vrrp[3174]: Netlink reflector reports IPfe80::20c:29ff:fe20:a6c8 addedOct 30 22:47:46node01 Keepalived_vrrp[3174]: Registering Kernel netlink reflectorOct 30 22:47:46node01 Keepalived_vrrp[3174]: Registering Kernel netlink command channelOct 30 22:47:46node01 Keepalived_vrrp[3174]: Registering gratuitous ARP shared channelOct 30 22:47:46node01 kernel: IPVS: Registered protocols (TCP, UDP, SCTP, AH, ESP)Oct 30 22:47:46node01 kernel: IPVS: Connection hash table configured (size=4096,memory=64Kbytes)Oct 30 22:47:46node01 kernel: IPVS: ipvs loaded.Oct 30 22:47:46node01 Keepalived_vrrp[3174]: Opening file '/etc/keepalived/keepalived.conf'.Oct 30 22:47:46node01 Keepalived_vrrp[3174]: Configuration is using : 63319 BytesOct 30 22:47:46node01 Keepalived_vrrp[3174]: Using LinkWatch kernel netlink reflector...Oct 30 22:47:46node01 Keepalived_healthcheckers[3173]: Interface queue is emptyOct 30 22:47:46node01 Keepalived_healthcheckers[3173]: Netlink reflector reports IP192.168.10.71 addedOct 30 22:47:46node01 Keepalived_healthcheckers[3173]: Netlink reflector reports IPfe80::20c:29ff:fe20:a6c8 addedOct 30 22:47:46node01 Keepalived_healthcheckers[3173]: Registering Kernel netlink reflectorOct 30 22:47:46node01 Keepalived_healthcheckers[3173]: Registering Kernel netlink command channelOct 30 22:47:46node01 Keepalived_healthcheckers[3173]: Opening file '/etc/keepalived/keepalived.conf'Oct 30 22:47:46node01 Keepalived_healthcheckers[3173]: Configuration is using : 11970 BytesOct 30 22:47:46node01 Keepalived_vrrp[3174]: VRRP sockpool: [ifindex(2), proto(112), fd(11,12)]Oct 30 22:47:46node01 Keepalived_healthcheckers[3173]: Using LinkWatch kernel netlink reflector...Oct 30 22:47:46node01 Keepalived_healthcheckers[3173]: Activating healthchecker for service[192.168.10.71]:3306Oct 30 22:47:46node01 kernel: IPVS: [wrr] scheduler registered.Oct 30 22:47:47node01 Keepalived_vrrp[3174]: VRRP_Instance(VI_1) Transition to MASTER STATEOct 30 22:47:48node01 Keepalived_vrrp[3174]: VRRP_Instance(VI_1) Entering MASTER STATEOct 30 22:47:48node01 Keepalived_vrrp[3174]: VRRP_Instance(VI_1) setting protocol VIPs.Oct 30 22:47:48node01 Keepalived_vrrp[3174]: VRRP_Instance(VI_1) Sending gratuitous ARPs oneth2 for 192.168.10.70Oct 30 22:47:48node01 Keepalived_healthcheckers[3173]: Netlink reflector reports IP192.168.10.70 added

8、观察node02的message日志，node02进入了BACKUP角色，VIP自然不会加上。

[root@node02 ~]# tail -f /var/log/messagesOct 30 22:48:54node02 Keepalived[16633]: Starting Keepalived v1.2.7 (09/26,2012)Oct 30 22:48:54node02 Keepalived[16634]: Starting Healthcheck child process, pid=16636Oct 30 22:48:54node02 Keepalived[16634]: Starting VRRP child process, pid=16637Oct 30 22:48:54node02 Keepalived_vrrp[16637]: Interface queue is emptyOct 30 22:48:54node02 Keepalived_healthcheckers[16636]: Interface queue is emptyOct 30 22:48:54node02 Keepalived_vrrp[16637]: Netlink reflector reports IP 192.168.10.72 addedOct 30 22:48:54node02 Keepalived_vrrp[16637]: Netlink reflector reports IPfe80::250:56ff:fe34:ca7 addedOct 30 22:48:54node02 Keepalived_vrrp[16637]: Registering Kernel netlink reflectorOct 30 22:48:54node02 Keepalived_vrrp[16637]: Registering Kernel netlink command channelOct 30 22:48:54node02 Keepalived_vrrp[16637]: Registering gratuitous ARP shared channelOct 30 22:48:54node02 Keepalived_vrrp[16637]: Opening file '/etc/keepalived/keepalived.conf'.Oct 30 22:48:54node02 Keepalived_healthcheckers[16636]: Netlink reflector reports IP192.168.10.72 addedOct 30 22:48:54node02 Keepalived_healthcheckers[16636]: Netlink reflector reports IPfe80::250:56ff:fe34:ca7 addedOct 30 22:48:54node02 Keepalived_healthcheckers[16636]: Registering Kernel netlink reflectorOct 30 22:48:54node02 Keepalived_healthcheckers[16636]: Registering Kernel netlink command channelOct 30 22:48:54node02 Keepalived_healthcheckers[16636]: Opening file'/etc/keepalived/keepalived.conf'.Oct 30 22:48:54node02 Keepalived_healthcheckers[16636]: Configuration is using : 11988 BytesOct 30 22:48:54node02 Keepalived_vrrp[16637]: Configuration is using : 63337 BytesOct 30 22:48:54node02 Keepalived_vrrp[16637]: Using LinkWatch kernel netlink reflector...Oct 30 22:48:54node02 Keepalived_vrrp[16637]: VRRP_Instance(VI_1) Entering BACKUP STATEOct 30 22:48:54node02 Keepalived_healthcheckers[16636]: Using LinkWatch kernel netlink reflector...Oct 30 22:48:54node02 Keepalived_vrrp[16637]: VRRP sockpool: [ifindex(2), proto(112),fd(10,11)]Oct 30 22:48:54node02 Keepalived_healthcheckers[16636]: Activating healthchecker forservice [192.168.10.72]:3306

9、测试vip是否可能用来登录mysql，由于此时vip在node01上，我们就在node02上使用vip来测试登录，可见是没有问题的

五、Mysql高可用测试

1、停止node01的mysql服务，3306端口自然不通，keepalived检测到3306端口不通后执行自杀脚本停止自身服务，VIP被移除释放出来

[root@node01 ~]# service mysql stopShutting down MySQL..                                     [  OK  ]

2、观察node01的messages日志可以明显看出整个过程，VIP也已经不见了

[root@node01 ~]# tail -f /var/log/messagesOct 30 23:06:42node01 Keepalived_healthcheckers[3173]: TCP connection to [192.168.10.71]:3306 failed !!!Oct 30 23:06:42node01 Keepalived_healthcheckers[3173]: Removing service [192.168.10.71]:3306from VS [192.168.10.70]:3306Oct 30 23:06:42node01 Keepalived_healthcheckers[3173]: Executing [/etc/keepalived/stopkeepalived.sh]for service [192.168.10.71]:3306 in VS [192.168.10.70]:3306Oct 30 23:06:42node01 Keepalived_healthcheckers[3173]: Lost quorum 1-0=1 > 0 for VS [192.168.10.70]:3306Oct 30 23:06:42node01 Keepalived_healthcheckers[3173]: SMTP connection ERROR to [192.168.200.1]:25.Oct 30 23:06:42node01 kernel: IPVS: __ip_vs_del_service: enterOct 30 23:06:42node01 Keepalived[3171]: Stopping Keepalived v1.2.7 (09/26,2012)Oct 30 23:06:42node01 Keepalived_vrrp[3174]: VRRP_Instance(VI_1) sending 0 priorityOct 30 23:06:42node01 Keepalived_vrrp[3174]: VRRP_Instance(VI_1) removing protocol VIPs.

3、观察node02的messages日志，可以看到node02进入了MASTER角色，接管了VIP，VIP已经加上，从日志的时间看，切换的过程不过花了1秒，可说是秒级切换了。

[root@node02 ~]# tail -f /var/log/messagesOct 30 23:06:42node02 Keepalived_vrrp[16637]: VRRP_Instance(VI_1) Transition to MASTER STATEOct 30 23:06:43node02 Keepalived_vrrp[16637]: VRRP_Instance(VI_1) Entering MASTER STATEOct 30 23:06:43node02 Keepalived_vrrp[16637]: VRRP_Instance(VI_1) setting protocol VIPs.Oct 30 23:06:43node02 Keepalived_vrrp[16637]: VRRP_Instance(VI_1) Sending gratuitous ARPson eth3 for 192.168.10.70Oct 30 23:06:43node02 Keepalived_healthcheckers[16636]: Netlink reflector reports IP 192.168.10.70 added

由于node01的keepalived进程被自杀脚本停止了，因此需要手动启动。之前我想是否需要跑一个监控脚本把keepalived服务自动开起来呢，后来我觉得不必要，因为如果mysql的服务依然异常，就算keepalived的服务起来了，它检测到本机的3306端口不通，还是会再次自杀。而既然mysql服务已经异常、端口都不通了，一般也是需要手动检查干预把mysql启动起来的，因此在mysql服务正常后再顺便手动起一下keepalived就好了。